Sorry, sent to the old mailing list address. -----Original Message----- From: Jennings, Richard R Sent: Tuesday, February 17, 2004 11:45 AM To: '[EMAIL PROTECTED]' Cc: 'RTnet Mailing List' Subject: RE: [RTnet-users] Question concerning RTMAC
Generally, the NIC handles collisions and retransmits at the hardware level unless there are a large number of transmit failures due to collisions. The number 16 comes to mind on the number handled in hardware. The Intel 82586/596 used only four bits to represent the "RETRY_NUM" in the configure command. The NIC starts transmission and if it detects another simultaneous transmission, it has detected a collision. It then stops the current transmission and does it's exponential (time) backoff algorithm to determine how long to wait before trying to transmit again. This is inherent in the CSMA/CD protocol used by Ethernet. If this backoff happens more than the number of times handled in hardware, a transmit error should be passed back to the software layer to handle it appropriately (either handled by the driver or TCP/IP stack). If you are not using rtmac, I would suggest you buy a full duplex switch to avoid collisions all together and only use hubs for a "monitoring tool". This will make your traffic much more deterministic (i.e. no backoffs due to collisions since backoffs can be a significant amount of time when we are talking "realtime"). HTH, Rich > -----Original Message----- > From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] > Sent: Tuesday, February 17, 2004 8:18 AM > To: [EMAIL PROTECTED] > Subject: [RTnet-users] Question concerning RTMAC > > > Hi everybody, > > I just play around with the 0.6.0 version of rtnet. > One thing is interesting - I think it is different from > previous versions: > I have a point-to-point (full duplex) connection from my > Intel eepro100 board (using the rtnet driver) and a non-PC > device. They communicate via UDP, this works fine. > However, when I do not use a point-to-point connection but > add a Hub to monitor the Ethernet traffic with a second PC, > the communication between the real time PC and the non-PC > device does not work as stable as it should work. > It looks, as if there are collisions that are not handled by > the system. > With an older version of rtnet (0.2.10) this seemed to work > fine as well. > > I do not use rtmac - as I think it is not required in my setup. > > My question is now: > If there are collisions on the Ethernet, is it the task of > the eepro100 hardware to do the detection of collisions and > re-sending of data or is this part of RTMac and this fails > in my setup as I do not use RTMac? > Does the usage of RTMac slow down the communication? > > Thanks for any feedback on that. > > Mathias > > > > -- > Mathias Koehrer > [EMAIL PROTECTED] ------------------------------------------------------- SF.Net is sponsored by: Speed Start Your Linux Apps Now. Build and deploy apps & Web services for Linux with a free DVD software kit from IBM. Click Now! http://ads.osdn.com/?ad_id56&alloc_id438&op=click _______________________________________________ RTnet-users mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/rtnet-users
